Artificial Analysis launches Intelligence Index v5, with 60% of scores from private test sets

What Artificial Analysis changed

Artificial Analysis published version 5 of its Intelligence Index on 27 October 2026. The company calls it the biggest upgrade since v4 launched in January. For the first time, most of the score comes from private data: 60% of the Index is now based on evaluations with private test sets or held-out answers, which makes it harder for model makers to train on the questions.

Two benchmarks are new. FrontierCode 2.0 is Cognition's private set of 132 coding tasks in real open-source repositories, graded on whether a maintainer would merge the change. Terminal-Bench-Science has 70 agentic research tasks run in a terminal. Three were upgraded: HLE-Diamond replaces Humanity's Last Exam, AA-Omniscience moves to v1.1, and CritPt v2 is scored on a private 40-question hard subset. GDPval-AA, SciCode and AA-LCR were removed. Knowledge work, coding, science and general knowledge each carry 25% of the weight.

How the leading models scored

At launch, Claude Opus 5.5 at max effort led with 58, followed by GPT-6 Astra at 57 and GPT-6.1 Sol at 55. Anthropic and OpenAI held the top five places, and Gemini 4 Argon scored 50. No model led all nine evaluations. Opus 5.5 led three, including FrontierCode 2.0 at 46%, and GPT-6 Astra led Terminal-Bench-Science at 63%.

The list changed within a day. On 28 October Artificial Analysis scored Gemini 4 Carbon at 59, first in the Index, with the top result in four of the nine evaluations. Google cited that ranking in its Carbon announcement. Artificial Analysis lists Carbon's cost at $7.76 per Index task at standard pricing, about 23% below Claude Opus 5.5 at max effort.

Open weights fall further behind

The new Index widens the gap between closed and open models. The best open weights entry, Kimi K3.1 Pro, scored 46, 12 points behind the leader, against a 3-point gap in the previous version. Its weights are not yet released. Among models with published weights, DeepSeek V4.1 Flash 1017 and GLM-5.3 scored 32, below Claude Haiku 5.5 at 36. On Terminal-Bench-Science most open weights models scored under 10%.

Cost and effort matter more too. Claude Opus 5.5 scores 35 at low effort and 58 at max effort, at ten times the cost per task. Among the top ten models, GPT-6.1 Terra at max effort was the cheapest at $0.96 per Index task. Every model produced more output tokens per task than in v4.3, and the median rose by two thirds.

What to watch next

Because most of v5 is private, outside researchers cannot rerun the full Index, so trust rests on the company's methodology page. Artificial Analysis also launched a Progress Index v1 that places models since November 2022 on one scale, built on lessons from Epoch AI, whose SpireClimb benchmark we covered this month. Expect labs to start citing v5 scores in release posts. Benchmark news is collected on the AI Models hub.

Questions people ask

Which model leads the Artificial Analysis Intelligence Index v5?

At launch on 27 October, Claude Opus 5.5 at max effort led with 58. On 28 October Gemini 4 Carbon took first place with 59, according to Artificial Analysis.

What is new in Intelligence Index v5?

Two new benchmarks, FrontierCode 2.0 and Terminal-Bench-Science, three upgraded ones, and 60% of the weight on private test sets or held-out answers.

How do open weights models do on Index v5?

The best open weights model, Kimi K3.1 Pro, scored 46, 12 points behind the leader. Among models with released weights, DeepSeek V4.1 Flash 1017 and GLM-5.3 led at 32.

Sources

1. Announcing Artificial Analysis Intelligence Index v5 (Artificial Analysis, 27 Oct 2026)
2. Gemini 4 Carbon is the new #1 on the Artificial Analysis Intelligence Index v5 (Artificial Analysis, 28 Oct 2026)